--- Log opened Wed Mar 13 00:00:11 2013 | ||
blackburn | I was in high school when house md first seasons were coming out | 00:01 |
---|---|---|
n4nd0 | yeah me too | 00:02 |
blackburn | as I was doing nothing in school I watched it | 00:02 |
n4nd0 | I watched house while in high school too | 00:02 |
blackburn | I finished school in 2008 IIRC | 00:03 |
blackburn | :D | 00:03 |
blackburn | n4nd0: when did you? | 00:03 |
n4nd0 | mmmmm | 00:03 |
n4nd0 | I finished high school 6 years ago | 00:03 |
blackburn | 2007 then I guess | 00:03 |
n4nd0 | 2007 probably | 00:03 |
n4nd0 | yeah | 00:04 |
n4nd0 | I started uni when I was 17 | 00:04 |
n4nd0 | now I am 23 | 00:04 |
n4nd0 | so yeah 2007, indeed | 00:04 |
blackburn | I am still 21 but 3 days lef | 00:04 |
blackburn | t | 00:04 |
n4nd0 | happy birthday!! | 00:05 |
blackburn | hah | 00:05 |
n4nd0 | haha | 00:05 |
blackburn | my brother has b-day today | 00:05 |
n4nd0 | happy birthday to him! | 00:05 |
blackburn | yeah I'll let him know, thanks :) | 00:05 |
n4nd0 | Saturday 16th yours right? | 00:05 |
blackburn | yeah | 00:05 |
blackburn | our parents were precise | 00:05 |
n4nd0 | haha yeah | 00:05 |
n4nd0 | older brother? | 00:06 |
blackburn | 13 and 16 march but 7 years | 00:06 |
blackburn | yes, 29 | 00:06 |
blackburn | n4nd0: https://dl.dropbox.com/u/10139213/share/image2.png | 00:09 |
blackburn | https://dl.dropbox.com/u/10139213/share/image.png | 00:09 |
blackburn | https://dl.dropbox.com/u/10139213/shogun/lle.png | 00:10 |
blackburn | n4nd0: ^ something like that in d3.js would be ideal | 00:10 |
n4nd0 | yeah | 00:10 |
n4nd0 | I have to see how to do it | 00:10 |
blackburn | https://dl.dropbox.com/u/10139213/shogun/pics.png | 00:10 |
blackburn | n4nd0: http://bl.ocks.org/mbostock/950642 | 00:12 |
blackburn | n4nd0: images as nodes | 00:13 |
n4nd0 | cool | 00:21 |
blackburn | n4nd0: http://bl.ocks.org/mbostock/1804919 that's an example of graph layout with 'suggested' position | 00:22 |
blackburn | n4nd0: so you could change gravity function, add image nodes and that's it | 00:23 |
blackburn | time to sleep | 00:24 |
n4nd0 | ok | 00:27 |
n4nd0 | gn! | 00:27 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 00:41 | |
-!- FSCV_ [~FSCV@65.19.131.247] has quit [Quit: Leaving] | 01:42 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 02:10 | |
shogun-buildbot | build #276 of nightly_none is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/276 | 03:01 |
-!- zxtx [~zv@cpe-98-148-115-57.socal.res.rr.com] has joined #shogun | 03:50 | |
-!- zxtx [~zv@cpe-98-148-115-57.socal.res.rr.com] has quit [Ping timeout: 258 seconds] | 05:48 | |
-!- blackburn [~blackburn@188.168.13.118] has quit [Ping timeout: 264 seconds] | 07:10 | |
-!- blackburn [~blackburn@188.168.3.95] has joined #shogun | 07:25 | |
-!- zxtx [~zv@76.91.87.166] has joined #shogun | 07:46 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 07:59 | |
shogun-notifier- | shogun: Sergey Lisitsyn :master * 3b6afdd / src/shogun/lib/tapkee/ (3 files): https://github.com/shogun-toolbox/shogun/commit/3b6afdd8ed5266523fff13bf0b061c944cac25c4 | 07:59 |
shogun-notifier- | shogun: Compilation fix for tapkee code which fails w/o ARPACK | 07:59 |
-!- zxtx [~zv@76.91.87.166] has quit [Ping timeout: 250 seconds] | 08:46 | |
-!- blackburn [~blackburn@188.168.3.95] has left #shogun [] | 08:51 | |
-!- n4nd0 [~nando@n177-p170.kthopen.kth.se] has joined #shogun | 08:59 | |
sonne|work | shogun-buildbot: force build nightly_none | 09:09 |
shogun-buildbot | build forced [ETA 15m26s] | 09:09 |
shogun-buildbot | I'll give a shout when the build finishes | 09:09 |
-!- blackburn [~lisitsin@mxs.kg.ru] has joined #shogun | 09:22 | |
shogun-buildbot | build #277 of nightly_none is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/277 | 09:24 |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 09:38 | |
-!- lambday [b613104d@gateway/web/freenode/ip.182.19.16.77] has joined #shogun | 10:05 | |
lambday | hi.. I've been writing a string kernel and I want to normalize the it... how do I do it? | 10:07 |
lambday | I've set the normalizer in the constructor as set_normalizer(new CSqrtDiagKernelNormalizer()); | 10:07 |
lambday | and tried to use return normalizer->normalize(total, idx_a, idx_b); in the compute function but got a segmentation fault | 10:08 |
blackburn | lambday: normalizer is being called in base CKernel class, no need to call it once more | 10:12 |
lambday | but compute function returns the unnormalized kernel value :( | 10:12 |
blackburn | lambday:https://github.com/shogun-toolbox/shogun/blob/master/src/shogun/kernel/Kernel.h#L234 | 10:16 |
blackburn | lambday: compute should return unnormalized value as kernel() computes final value | 10:16 |
lambday | blackburn: oh... I get it now.. | 10:17 |
lambday | thanks a lot | 10:17 |
sonne|work | lambday: just make sure that you call init_normalizer() in kernels' init(a,b) function | 10:25 |
lambday | sonne|work: yes... I did and its working fine now :) | 10:26 |
sonne|work | k | 10:26 |
lambday | one more thing... I want to add this SSKStringKernel to modular interface... what exactly should I do? blackburn told me few days back that it needs to be added to interfaces/modular/Kernel.i, and Kernel_includes.i. | 10:26 |
lambday | I see a few renames and includes in Kernel.i | 10:27 |
blackburn | lambday: that's the only thing | 10:27 |
blackburn | yes, rename and include in Kernel.i | 10:27 |
blackburn | and just include in Kernel_includes.i | 10:27 |
lambday | blackburn: okay... i'm checking | 10:28 |
-!- Netsplit *.net <-> *.split quits: sonne|work | 10:36 | |
-!- Netsplit over, joins: sonne|work | 10:46 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 10:59 | |
lambday | blackburn: thanks man... its working perfectly... the kernel matrix is symmetric.. tested with some data.. | 11:32 |
sonne|work | lambday: now it only has to be positive :-] | 11:33 |
lambday | how do i test its psd? | 11:33 |
sonne|work | lambday: empirically? check if eigenvalues are positive | 11:36 |
lambday | sonne|work: I gotta check it manually? :( | 11:39 |
sonne|work | lambday: yeah proof it! | 11:40 |
lambday | sonne|work: okay... | 11:40 |
sonne|work | lambday: but don't worry if it is not - we have some 'kernels' in there that are not by theory but empirically everything is good | 11:42 |
lambday | sonne|work: that sounds relaxing :) | 11:45 |
lambday | I gotta check its performance on some protein test data... I should compare its performance against other string kernels... let's see :) | 11:47 |
lambday | thanks a lot :) | 11:47 |
-!- blackburn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 11:58 | |
-!- blackburn [~lisitsin@mxs.kg.ru] has joined #shogun | 12:20 | |
-!- heiko [~heiko@pat-191-250.internal.eduroam.ucl.ac.uk] has joined #shogun | 12:35 | |
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has joined #shogun | 12:36 | |
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has quit [Ping timeout: 245 seconds] | 12:47 | |
blackburn | heiko: I have fixed something | 12:50 |
heiko | blackburn: nice :)( | 12:50 |
heiko | I am scanning the gpr framework | 12:51 |
heiko | found some bugs | 12:51 |
heiko | still results are nonsense | 12:51 |
heiko | hope to finish this today | 12:51 |
heiko | at least for regression | 12:51 |
heiko | btw while looking through this, I think I will write a project description for GP classification for gsoc | 12:51 |
heiko | spent whole yesterday evening with this gpr stuff, we really have to make people write tests | 12:53 |
blackburn | heiko: I don't get a thing about GP - that's a lot of code I am quite lost within | 12:57 |
heiko | blackburn: I know it should have been better documented/organised | 12:58 |
heiko | the variable names are very misleading | 12:58 |
heiko | but I am on it | 12:58 |
blackburn | heiko: that's "-" of part-time gsoc development | 12:58 |
blackburn | "+" is that we get some code though | 12:58 |
heiko | blackburn: I think that is our fault | 12:58 |
heiko | not the student's | 12:58 |
heiko | since we did not make them | 12:58 |
heiko | so lets do this differently this year | 12:58 |
blackburn | heiko: exactly | 12:59 |
heiko | I wont merge any new code without a unit test | 12:59 |
blackburn | heiko: that's always management | 12:59 |
heiko | and without loads of comments | 12:59 |
blackburn | heiko: that's actually my PM who should be blamed I am reading about C++11 instead of doing something useful :D | 12:59 |
blackburn | so yes, I agree | 12:59 |
blackburn | heiko: it would be nice if you declare such principle on the mailing list | 13:00 |
blackburn | I could send a scan of my sign :D | 13:00 |
blackburn | heiko: there is an anecdote about russian diplomates | 13:01 |
heiko | blackburn: yes? :) | 13:01 |
blackburn | heiko: something like that: teachers asked future diplomates to write a letter to somalian pirates | 13:01 |
blackburn | which would force to set hostages free | 13:01 |
heiko | haha :) | 13:02 |
blackburn | week later - teacher "nice, but assholes should be written without a space and you made some mistakes in word f*cking" | 13:02 |
blackburn | heiko: that's a lame translation but I hope you get it right :) | 13:03 |
heiko | I get it | 13:03 |
heiko | well, we have the power to merge :) | 13:03 |
blackburn | heiko: in russian it sounds funnier as we have infinite amounts of various blames | 13:04 |
blackburn | heiko: so when you have some time and will, could you please write a manifest about that? | 13:05 |
heiko | blackburn: I will, we can put it on the GSoC page | 13:05 |
heiko | like guielines | 13:05 |
heiko | on developing code | 13:05 |
blackburn | heiko: yeap | 13:06 |
heiko | blackburn: yeah I fixed the bug | 13:12 |
heiko | blackburn: would have been easily catched by a test | 13:12 |
heiko | caused by something that was developed after this first thing was checked | 13:12 |
heiko | #but since not test, no detection that old breaks when new is added | 13:12 |
blackburn | heiko: that's quite cool we all get importance of unit-tests | 13:13 |
heiko | blackburn: sorry :) I am too keen on this | 13:13 |
blackburn | heiko: no that's ok | 13:13 |
blackburn | heiko: ideal case is first tests then code | 13:15 |
heiko | blackburn: yes, code a module/method, add test, then use it in code | 13:15 |
blackburn | heiko: it won't work with software you don't know how to write but for established stuff it is the only way | 13:15 |
heiko | blackburn: well ideally one things of the stuff you gonna write before you do it right? :) | 13:16 |
blackburn | heiko: well interface should be know first - so interface tests should come before real coding | 13:16 |
blackburn | (ideal) | 13:16 |
blackburn | heiko: one thing sklearn rocks with is that they have ->100% test coverage | 13:17 |
heiko | i know | 13:18 |
heiko | but we will increase ours ! | 13:19 |
-!- n4nd0 [~nando@n177-p170.kthopen.kth.se] has quit [Ping timeout: 246 seconds] | 13:45 | |
sonne|work | heiko: please don't document code! | 14:17 |
sonne|work | (I am serious here) | 14:17 |
heiko | sonne|work: not doing | 14:17 |
blackburn | sonne|work: that's interesting - motivate ;) | 14:17 |
heiko | fixing | 14:17 |
sonne|work | comments are not a good idea | 14:18 |
heiko | ? | 14:18 |
sonne|work | only for doxygen api use | 14:18 |
blackburn | yeah that's true, code should be selfcommenting | 14:18 |
sonne|work | heiko: if you need comments your code is not readable | 14:18 |
heiko | sonne|work: I do not agree | 14:18 |
heiko | I think it is very good to explain what one is doing | 14:18 |
sonne|work | heiko: no | 14:18 |
sonne|work | heiko: give the function a proper name | 14:18 |
sonne|work | instead of calling it do_stuff | 14:19 |
sonne|work | call it compute_variance etc | 14:19 |
sonne|work | -> no comment necessary | 14:19 |
heiko | its way easier to read "compute cholesky" than | 14:19 |
heiko | cblas_dsymm(CblasColMajor, CblasLeft, CblasUpper, | 14:19 |
heiko | m_ktrtr.num_rows, temp2.num_cols, (m_scale*m_scale)/(sigma*sigma), | 14:19 |
heiko | m_ktrtr.matrix, m_ktrtr.num_cols, | 14:19 |
heiko | temp2.matrix, temp2.num_cols, 1.0, | 14:19 |
heiko | temp1.matrix, temp1.num_cols); | 14:19 |
heiko | memcpy(m_kern_with_noise.matrix, temp1.matrix, | 14:19 |
heiko | temp1.num_cols*temp1.num_rows*sizeof(float64_t)); | 14:19 |
heiko | clapack_dpotrf(CblasColMajor, CblasUpper, | 14:19 |
heiko | temp1.num_rows, temp1.matrix, temp1.num_cols); | 14:19 |
heiko | m_L = SGMatrix<float64_t>(temp1.num_rows, temp1.num_cols); | 14:19 |
blackburn | heiko: eigen3 :) | 14:19 |
heiko | blackburn: I did not write that! :) | 14:19 |
blackburn | true | 14:19 |
sonne|work | heiko: yes but call it compute_cholesky | 14:19 |
sonne|work | and put it in a function | 14:20 |
heiko | sonne|work: also sometimes its good to describe what the next section will be doing | 14:20 |
sonne|work | no code needed | 14:20 |
sonne|work | heiko: no! | 14:20 |
sonne|work | exactly not! | 14:20 |
sonne|work | if you want to comment what the next section does | 14:20 |
blackburn | heiko: matrix.ldlt().solve(rhs) | 14:20 |
sonne|work | put it in a function and give it a self-describing name! | 14:20 |
blackburn | heiko: ^ eigen3 way | 14:20 |
sonne|work | heiko: btw what is the state about this GP stuff? | 14:20 |
heiko | sonne|work: I do not agree, of course this is what one should do, but still, sometimes comments are needed. | 14:21 |
heiko | sonne|work, I solved the bug that made the basic GPR return nonsense | 14:21 |
sonne|work | very extremely rarely | 14:21 |
sonne|work | nice! | 14:21 |
heiko | and wrote many tests to check | 14:21 |
heiko | but now | 14:21 |
heiko | it crashes depending on size of test data | 14:21 |
heiko | (it did that before) | 14:21 |
heiko | so investigating .... | 14:21 |
heiko | at least the regression works if it terminates | 14:22 |
heiko | blackburn: I know, but I am not going to change all these calls :) (as long as they work) | 14:23 |
blackburn | heiko: just persuading you to join the dark side | 14:24 |
heiko | hey, I use eigen3 for my current project! :) | 14:24 |
heiko | btw exactly this call you mentioned :) | 14:24 |
blackburn | heiko: then I am helping you with motivation to not left the dark side :) | 14:25 |
heiko | hehe | 14:25 |
heiko | blackburn: maybe I should use eigen3, there is an error in the cholesky for certain matrix sizes | 14:27 |
heiko | blackburn, whats the best way to convert a shogun matrix into an eigen3 one? | 14:28 |
heiko | I think thats the problem, there is some uninitialised memory | 14:29 |
blackburn | heiko: Map class | 14:29 |
blackburn | heiko: Map<MatrixXd>(pointer,cols,rows) | 14:30 |
blackburn | heiko: act just like with typical matrix afterwards | 14:31 |
heiko | Map class of eigen3? | 14:31 |
heiko | so this returns a 2d array? | 14:31 |
blackburn | heiko: yes, just like MatrixXf | 14:33 |
blackburn | or MatrixXd | 14:33 |
heiko | nice, what about K^T * x , blackburn? | 14:39 |
heiko | how do I do that? | 14:39 |
heiko | will get lunch now, see you in a bit | 14:40 |
blackburn | heiko: K.transpose()*x | 14:43 |
-!- n4nd0 [~nando@n177-p170.kthopen.kth.se] has joined #shogun | 15:01 | |
-!- FSCV [~FSCV@204.45.132.11] has joined #shogun | 15:02 | |
heiko | blackburn, does that work in place? | 15:09 |
heiko | I guess so | 15:09 |
blackburn | heiko: no, y = K*transpose()*x will form it inplcae | 15:12 |
heiko | blackburn: I see thanks | 15:12 |
blackburn | but just K.transpose()*x does nothing | 15:12 |
heiko | blackburn, LLT | 15:13 |
blackburn | heiko: mutable stuff is in array mainly | 15:13 |
heiko | does it change the given matrix? | 15:13 |
blackburn | heiko: solveInPlace() | 15:13 |
blackburn | I think you need it if you want to change it | 15:13 |
heiko | blackburn: is there any example eigen code in shogun that I could borrow some things from? | 15:16 |
blackburn | heiko: argh wait that's not true | 15:16 |
blackburn | heiko: LLT computes a copy | 15:16 |
blackburn | yes | 15:16 |
blackburn | heiko: https://github.com/shogun-toolbox/shogun/tree/master/src/shogun/lib/malsar | 15:16 |
blackburn | these algorithms are made of eigen | 15:17 |
heiko | Map<VectorXd> Mzp_DiagSigz(x,n_tasks);Mzp_Pz = eigensolver.eigenvectors().real(); | 15:17 |
blackburn | heiko: yeah I had to keep naming to track with original implementation | 15:18 |
heiko | no worries, just asking whether this map is then the 2d array? | 15:18 |
blackburn | heiko: yes, a vector of variable length (X) and (d)ouble type | 15:19 |
blackburn | heiko: this code adhere Mzp_DiagSigz to pointer x | 15:19 |
blackburn | and then steal eigenvectors I guess | 15:19 |
blackburn | I don't remember exactly how can that work though | 15:20 |
blackburn | :D | 15:20 |
blackburn | eigenvectors must be a matrix.. | 15:20 |
heiko | blackburn: | 15:21 |
heiko | sorry I always have problems with new frameworks | 15:22 |
heiko | what is the code | 15:22 |
heiko | I have a shogun matrix , I want to turn it into MatrixXd | 15:22 |
heiko | or similar | 15:22 |
blackburn | heiko: map is the thing you need | 15:22 |
heiko | how | 15:23 |
heiko | Map<MatrixXd> K(m_kern_with_noise.matrix, m_ktrtr.num_rows, | 15:23 |
heiko | m_ktrtr.num_cols); | 15:23 |
heiko | I did this | 15:23 |
heiko | now I want to add something to the diagonal, how do I do that? | 15:23 |
heiko | how to get the MatrixXd instance? | 15:23 |
heiko | blackburn: ah I finally found the documentation site :) nevermind | 15:24 |
-!- FSCV [~FSCV@204.45.132.11] has quit [Ping timeout: 245 seconds] | 15:25 | |
blackburn | heiko: sorry getting distracted | 15:27 |
blackburn | heiko: asDiagonal() += vec | 15:27 |
blackburn | K.asDiagonal().array() += vec | 15:27 |
blackburn | sth like that IIRC | 15:27 |
blackburn | heiko: but actually yes they have not that bad doc | 15:28 |
blackburn | some things are not clear though | 15:28 |
blackburn | I asked about some thing once on their channel | 15:28 |
blackburn | the answer I got was 'check doc' | 15:28 |
blackburn | go figure :D | 15:28 |
heiko | ha | 15:28 |
heiko | so now I have a MatrixXd | 15:30 |
heiko | how to get the 2d array? :D | 15:30 |
blackburn | heiko: what do you mean by array? | 15:32 |
sonne|work | blackburn: double[] | 15:32 |
heiko | yes | 15:32 |
blackburn | heiko: K.data() | 15:32 |
heiko | thanks! | 15:33 |
blackburn | that's pointer for me | 15:33 |
blackburn | :D | 15:33 |
heiko | blackburn, what I am trying to do: | 15:34 |
heiko | I have a shogun matrix | 15:34 |
heiko | and I want another shogun matrix of the cholesky of it | 15:34 |
heiko | so it seems like I have to copy the data twice | 15:34 |
heiko | or once? | 15:34 |
heiko | or at all=? | 15:34 |
blackburn | heiko: first map | 15:36 |
blackburn | existing matrix | 15:36 |
blackburn | you get Map<MatrixXd> | 15:36 |
blackburn | then prepare llt() | 15:36 |
blackburn | ahh I see what you mean | 15:36 |
blackburn | yes matrix would try to shoot your leg if you just steal the matrix | 15:37 |
blackburn | * the array | 15:37 |
heiko | yes, I need to store the llt suff | 15:37 |
heiko | so I need to memcpy right? | 15:38 |
blackburn | heiko: yes I think so | 15:38 |
blackburn | heiko: but I'd avoid memcpy here | 15:38 |
heiko | so what to do? | 15:39 |
blackburn | heiko: I don't like that solution | 15:39 |
blackburn | heiko: just copy values with for | 15:39 |
blackburn | heiko: but again that feels wrong, let me check | 15:40 |
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has joined #shogun | 15:40 | |
blackburn | heiko: why do you need it btw? | 15:41 |
heiko | blackburn: because the GP code stored the L for late ruse | 15:41 |
blackburn | can't you store LLT? | 15:41 |
blackburn | ahh | 15:41 |
heiko | and I dont want to store an eigen object | 15:41 |
blackburn | heiko: alright, copying is quite bad though | 15:42 |
heiko | why? isnt it faster than for | 15:42 |
blackburn | heiko: I mean in general I don't like that you've got to copy it twice | 15:43 |
heiko | it is only once right? | 15:43 |
heiko | the map should not copy? | 15:43 |
blackburn | heiko: I think I got a solution | 15:43 |
heiko | or does llt copy? | 15:43 |
blackburn | okay | 15:43 |
blackburn | you have SGMatrix G | 15:43 |
heiko | lets call it K | 15:43 |
blackburn | you create Map<MatrixXd> K | 15:43 |
blackburn | okay | 15:43 |
blackburn | sgK | 15:43 |
blackburn | and eigK | 15:43 |
blackburn | you also have matrix SGMatrix L | 15:44 |
heiko | eigK is the ma?p | 15:44 |
blackburn | which is cholesky factor | 15:44 |
blackburn | you add Map<MatrixXd> eigL | 15:44 |
blackburn | and | 15:44 |
blackburn | eigL = eigK.llt().matrixL(); | 15:44 |
heiko | I see | 15:44 |
heiko | good idea | 15:44 |
heiko | btw how to I get upper? | 15:44 |
blackburn | heiko: that sounds like a most efficient way to do that | 15:44 |
heiko | like in matlab? | 15:44 |
blackburn | heiko: matrixU but what is upper? | 15:44 |
blackburn | do you need it? | 15:45 |
heiko | the upper triangular part | 15:45 |
heiko | yes | 15:45 |
blackburn | heiko: alright then matrixL and matrixU | 15:45 |
heiko | ok | 15:45 |
heiko | thanks | 15:45 |
heiko | I will try ... | 15:45 |
blackburn | heiko: these matrices are actually a views | 15:45 |
blackburn | heiko: ah btw if your K is symmetric | 15:46 |
blackburn | K.selfadjointView<Eigen::Upper>().llt().matrixU() | 15:47 |
blackburn | that can be more efficient | 15:47 |
-!- FSCV [~FSCV@vpnreactor-static44.den.icastcenter.com] has joined #shogun | 15:47 | |
heiko | what does it do? | 15:48 |
blackburn | heiko: it tells eigen that your matrix can be treated as selfadjoint | 15:49 |
blackburn | thus all operations that can improve from that would improve from that | 15:50 |
blackburn | heiko: for example I use it in PCA-like algorithms | 15:50 |
blackburn | covariance is C.selfadjointView<Eigen::Upper>().rankUpdate(v,v,1.0); | 15:50 |
blackburn | rank update is C <- C + alpha*x*x' - it is done much faster then you know C is symmetric | 15:51 |
blackburn | when* | 15:51 |
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has quit [Ping timeout: 245 seconds] | 15:55 | |
heiko | blackburn: ha you know what | 16:03 |
heiko | the cholesky is stored | 16:03 |
heiko | but not used for solving | 16:03 |
blackburn | heiko: are you talking about the bug? | 16:06 |
heiko | blackburn: plural, but no | 16:06 |
heiko | the thing about storing the factor | 16:06 |
heiko | say is there a way to solve a system with an existing cholesky? | 16:06 |
heiko | llt.solve(b) | 16:06 |
heiko | only does it for the computed one | 16:07 |
blackburn | heiko: ahh yes that's a problem | 16:07 |
blackburn | heiko: they didn't expect such thing I guess :) | 16:09 |
heiko | blackburn: mmh that is annoying | 16:09 |
heiko | I mean | 16:09 |
heiko | currently the GP framework solves again anyway | 16:09 |
heiko | but it would be good not to do that | 16:09 |
heiko | uuuh | 16:10 |
heiko | there are so many bugs | 16:11 |
heiko | its not possible to deal with all of them | 16:11 |
blackburn | heiko: okay I'll get back to you in a hour or a little more | 16:11 |
heiko | ok see you | 16:11 |
blackburn | switching place | 16:11 |
-!- FSCV_ [~FSCV@85.17.20.88] has joined #shogun | 16:14 | |
-!- blackburn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] | 16:14 | |
-!- FSCV [~FSCV@vpnreactor-static44.den.icastcenter.com] has quit [Ping timeout: 248 seconds] | 16:16 | |
-!- FSCV_ [~FSCV@85.17.20.88] has quit [Ping timeout: 260 seconds] | 16:38 | |
-!- FSCV_ [~FSCV@76.73.18.131] has joined #shogun | 16:38 | |
-!- FSCV_ [~FSCV@76.73.18.131] has quit [Ping timeout: 264 seconds] | 16:46 | |
-!- FSCV_ [~FSCV@65.19.131.247] has joined #shogun | 16:51 | |
-!- n4nd0 [~nando@n177-p170.kthopen.kth.se] has quit [Quit: leaving] | 17:08 | |
-!- blackburn [~blackburn@83.234.54.160] has joined #shogun | 17:11 | |
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has quit [Quit: Leaving.] | 17:48 | |
blackburn | heiko: I'll have some 2-3 hours tonight for shogun stuff I think, let me know what can I fix | 17:54 |
heiko | blackburn: cool will do | 18:07 |
heiko | gpr is about to be working again | 18:07 |
heiko | just had a break | 18:07 |
heiko | blackburn: heureca, it works! | 18:31 |
heiko | blackburn: multiclass ocas test fails | 18:45 |
blackburn | heiko: alright | 18:49 |
blackburn | heiko: mc ocas then | 18:49 |
blackburn | heiko: what was wrong with gps? | 18:51 |
heiko | blackburn: two things: | 18:51 |
heiko | a wrong feature object was used at a place so the thing was doing a different thing that it should have | 18:52 |
heiko | and a lapack method was wrongly called, causing memory errors/crashes | 18:52 |
heiko | seems to work quite nice now | 18:52 |
heiko | but only tested the basic GPR | 18:52 |
blackburn | I am going to update tapkee in shougn now | 18:52 |
heiko | which is only a small part of the framework | 18:52 |
blackburn | and then check mc ocas | 18:52 |
heiko | blackburn: no | 18:52 |
heiko | lets do bugs | 18:52 |
heiko | I suggest | 18:52 |
blackburn | heiko: that's just one commit | 18:52 |
heiko | blackburn: ok :) | 18:52 |
blackburn | heiko: I feel uncomfortable when tapkee is out of date in shogun | 18:53 |
blackburn | heiko: I had to write a script that updates all includes to match that <shogun/lib/tapkee/...> :D | 18:54 |
heiko | here we go! | 18:56 |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 18:57 | |
shogun-notifier- | shogun: Heiko Strathmann :master * 1a024d4 / / (5 files): https://github.com/shogun-toolbox/shogun/commit/1a024d413ab76903a24c0876f12b9e2bbd15be99 | 18:57 |
shogun-notifier- | shogun: added additional parameter to mean shift data generator (select dimension of shift) and make sure that tests/examples still work | 18:57 |
shogun-notifier- | shogun: Heiko Strathmann :master * b847c03 / examples/undocumented/python_modular/library_time.py: https://github.com/shogun-toolbox/shogun/commit/b847c03973f7855bb5a3f873a6ce081aebe98e8e | 18:57 |
shogun-notifier- | shogun: added more test cases and make use of parameter list | 18:57 |
shogun-notifier- | shogun: Heiko Strathmann :master * 7c25b13 / src/shogun/regression/ (2 files): https://github.com/shogun-toolbox/shogun/commit/7c25b13b5d880ea87aa89f28a3c3e42850300855 | 18:57 |
shogun-notifier- | shogun: fixes the bug that GP regression returns nonsense | 18:57 |
shogun-notifier- | shogun: Heiko Strathmann :master * f37c2d8 / src/shogun/regression/gp/InferenceMethod.h: https://github.com/shogun-toolbox/shogun/commit/f37c2d87321ab9dc71c77a8d615343d4510c6ae6 | 18:57 |
shogun-notifier- | shogun: added a note | 18:57 |
shogun-notifier- | shogun: Heiko Strathmann :master * 8bd57ff / src/shogun/regression/gp/ExactInferenceMethod.cpp: https://github.com/shogun-toolbox/shogun/commit/8bd57ff2b969f63f66f29f8ddc7b9c039a33c262 | 18:57 |
shogun-notifier- | shogun: replaced lapack by eigen3 calls and make code much easier to read | 18:57 |
shogun-notifier- | shogun: Heiko Strathmann :master * 58366b5 / examples/undocumented/libshogun/ (2 files): https://github.com/shogun-toolbox/shogun/commit/58366b5899ed5b890fd27ca379780023bd8a296c | 18:57 |
shogun-notifier- | shogun: added a simply GP regression example | 18:57 |
shogun-notifier- | shogun: Heiko Strathmann :master * 42a7929 / tests/unit/regression/ (2 files): https://github.com/shogun-toolbox/shogun/commit/42a79292ec8311cc2c8f4a4bb94e02189a86e176 | 18:57 |
shogun-notifier- | shogun: new unit tests for gp regression with exact inference | 18:57 |
shogun-notifier- | shogun: Merge pull request #919 from karlnapf/master | 18:57 |
shogun-notifier- | shogun: | 18:57 |
shogun-notifier- | shogun: Made classic GP regression work again | 18:57 |
blackburn | heiko: big thin | 18:57 |
blackburn | g | 18:57 |
heiko | blackburn: when you fix ocas, also update the integration test since it is broken | 18:58 |
heiko | blackburn: yes, that was a nasty one | 18:58 |
blackburn | alright | 18:58 |
heiko | eigen3 makes things much easier btw | 18:58 |
heiko | 30 lines -> 4 lines | 18:58 |
heiko | and no bugs | 18:58 |
blackburn | heiko: you get it | 18:58 |
heiko | blackburn: converter_multidimensionalscaling_modular.py also has a broken integration test, could you add unit test to make sure it works and then replace integration test? | 18:59 |
blackburn | heiko: yeah I changed it and will replace integration test | 18:59 |
-!- lambday [b613104d@gateway/web/freenode/ip.182.19.16.77] has quit [Quit: Page closed] | 19:01 | |
blackburn | white smoke! | 19:10 |
-!- AnkitAgrawal [3ba21749@gateway/web/freenode/ip.59.162.23.73] has joined #shogun | 19:13 | |
-!- AnkitAgrawal [3ba21749@gateway/web/freenode/ip.59.162.23.73] has quit [Ping timeout: 245 seconds] | 19:31 | |
blackburn | heiko: I like the system I built in tapkee | 19:38 |
blackburn | terminate called after throwing an instance of 'tapkee::wrong_parameter_type_error' | 19:38 |
blackburn | what(): Wrong type for parameter target_dimension. Should be IndexType | 19:38 |
blackburn | heiko: quite easy to fix anything :) | 19:39 |
heiko | blackburn | 19:42 |
heiko | multiclass labels cannot be serialized | 19:42 |
heiko | due to SGVector<SGVector ... | 19:42 |
heiko | could you replace that by a SGMatrix? | 19:42 |
blackburn | heiko: oops | 19:42 |
heiko | number is known | 19:42 |
blackburn | yeah | 19:42 |
blackburn | yes sure | 19:42 |
heiko | cool, let me know when its done, I have more integration tests working then :) | 19:43 |
blackburn | heiko: how many are failing? | 19:43 |
heiko | not too many | 19:43 |
heiko | getting there | 19:43 |
heiko | let me check | 19:43 |
blackburn | heiko: we won't get it to zero though | 19:43 |
heiko | blackburn: I know | 19:43 |
heiko | but we fix bugs | 19:44 |
heiko | that is important | 19:44 |
heiko | more importantly: yet unknown bugs | 19:44 |
heiko | the known ones are in the giuthub issue list | 19:44 |
heiko | so everyone can check | 19:44 |
heiko | blackburn: btw I really like this matrix view of eigen | 19:45 |
blackburn | heiko: which one? | 19:45 |
blackburn | selfadjoint? | 19:45 |
heiko | no the map in general | 19:45 |
blackburn | ah | 19:45 |
heiko | thats cool stuff | 19:46 |
heiko | blackburn: I will go home now, need some rest, worked all day on this stuff | 19:46 |
blackburn | heiko: people have understood how to C++ finally | 19:46 |
blackburn | :D | 19:46 |
heiko | blackburn: yeah :) | 19:46 |
blackburn | yeah have a rest, I am on the duty now | 19:46 |
heiko | after killing themselves a few times in the future :D | 19:46 |
heiko | blackburn: I have some time tomorrow so I can do final things | 19:47 |
blackburn | exactly | 19:47 |
heiko | when to release tomorrow? | 19:47 |
heiko | evening? | 19:47 |
blackburn | heiko: I don't know | 19:47 |
blackburn | the master should say | 19:47 |
heiko | ok then | 19:47 |
heiko | so see you later or tomorrow more probably | 19:47 |
heiko | let me know when the multiclass labels are changed | 19:48 |
blackburn | heiko: I am writing tests for tapkee | 19:48 |
heiko | I will change the test the | 19:48 |
heiko | n | 19:48 |
blackburn | I am not sure if I should include it too | 19:48 |
blackburn | in shogun | 19:48 |
heiko | what? | 19:48 |
heiko | tests? | 19:48 |
blackburn | yeah | 19:48 |
heiko | dont know | 19:48 |
blackburn | heiko: yeah me too | 19:48 |
heiko | is it that distinct? | 19:48 |
blackburn | the code of tapkee is just the same | 19:49 |
blackburn | in standalone and in shogun | 19:49 |
blackburn | so in principle they are not really required | 19:49 |
heiko | I would say yes since then we can see them from shogun | 19:49 |
blackburn | but tests for interface of tapkee in shogun (CClasses) should be here for sure | 19:49 |
heiko | you know, I dont have to get involved into yet another framework then :D | 19:49 |
blackburn | heiko: heh I use gtest too | 19:50 |
blackburn | so I can just put them directly.. | 19:50 |
blackburn | heiko: almost no alternative to gtest for C++ I think | 19:51 |
blackburn | boost.test may be | 19:51 |
heiko | dont know | 19:51 |
heiko | blackburn: have a good evening, see you around | 19:51 |
blackburn | see you | 19:51 |
@sonney2k | heiko, afternoon | 19:51 |
@sonney2k | tomorrow would be good | 19:51 |
blackburn | sonney2k: tomorrow evening? | 19:52 |
@sonney2k | afternoon | 19:52 |
blackburn | alright | 19:53 |
heiko | sonney2k: ok then | 19:57 |
heiko | sonney2k: why not evening? | 19:57 |
heiko | sonney2k: I have some more time in evening not so much in the afternoon (even earlier here) | 19:57 |
blackburn | libshogun.a 280 mb | 19:58 |
blackburn | hah | 19:58 |
@sonney2k | heiko, the release process takes hours... and I want to get some sleep.... | 20:03 |
heiko | sonney2k: okay, thats a good point | 20:04 |
heiko | :) | 20:04 |
-!- heiko [~heiko@pat-191-250.internal.eduroam.ucl.ac.uk] has left #shogun [] | 20:06 | |
shogun-notifier- | shogun: Sergey Lisitsyn :master * d3d518c / src/shogun/lib/tapkee/ (25 files): https://github.com/shogun-toolbox/shogun/commit/d3d518cacac72cf33d7078be3063250c35293c2d | 20:35 |
shogun-notifier- | shogun: A big update for Tapkee | 20:35 |
shogun-notifier- | shogun: | 20:35 |
shogun-notifier- | shogun: - Fixed a few VS 2012 compilations | 20:35 |
shogun-notifier- | shogun: - Parallel MDS | 20:35 |
shogun-notifier- | shogun: - Parallel | 20:35 |
shogun-buildbot | build #959 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/959 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 20:38 |
blackburn | damn | 20:38 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has joined #shogun | 20:39 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/5480717 | 20:39 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has left #shogun [] | 20:39 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 20:47 | |
shogun-notifier- | shogun: Sergey Lisitsyn :master * 6e01346 / src/shogun/lib/tapkee/tapkee_shogun.cpp: https://github.com/shogun-toolbox/shogun/commit/6e013467d73a06c61fa89c808eaa06927af546e3 | 20:59 |
shogun-notifier- | shogun: Fixed compilation | 20:59 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has joined #shogun | 21:06 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/5481238 | 21:06 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has left #shogun [] | 21:06 | |
shogun-buildbot | build #960 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/960 | 21:10 |
shogun-buildbot | build #772 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/772 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 21:16 |
shogun-notifier- | shogun: Sergey Lisitsyn :master * 48366db / / (4 files): https://github.com/shogun-toolbox/shogun/commit/48366db1a2f832edcae22d05382a91fb4e5bef0d | 22:03 |
shogun-notifier- | shogun: Some improvement for multiclass confidences | 22:03 |
shogun-notifier- | shogun: | 22:03 |
shogun-notifier- | shogun: - Changed the way confidences are stored (array of sgvectors -> sgmatrix) | 22:03 |
shogun-notifier- | shogun: | 22:03 |
shogun-notifier- | shogun: - Added a unit test for confidences | 22:03 |
blackburn | oops | 22:06 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has joined #shogun | 22:11 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/5482719 | 22:11 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has left #shogun [] | 22:11 | |
shogun-buildbot | build #787 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/787 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 22:18 |
blackburn | bad attempt haha | 22:19 |
shogun-buildbot | build #773 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/773 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 22:21 |
shogun-notifier- | shogun: Sergey Lisitsyn :master * c80ac5e / src/shogun/machine/MulticlassMachine.cpp: https://github.com/shogun-toolbox/shogun/commit/c80ac5e045d35c0d37d857e7b3dcec5a926cd7e2 | 22:21 |
shogun-notifier- | shogun: Fixed MC machine confidences allocation | 22:21 |
shogun-buildbot | build #774 of bsd1 - libshogun is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/774 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 22:29 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has joined #shogun | 22:30 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/5483214 | 22:30 |
-!- travis-ci [~travis-ci@ec2-50-19-78-12.compute-1.amazonaws.com] has left #shogun [] | 22:30 | |
shogun-buildbot | build #912 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/912 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 22:54 |
shogun-buildbot | build #788 of deb2 - static_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/788 | 23:06 |
shogun-buildbot | build #913 of deb3 - modular_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/913 | 23:11 |
--- Log closed Thu Mar 14 00:00:12 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!